Survey of Camera Error Sources in Machine Vision Systems.

摘要

In machine vision applications, such as an automated inspection line, television cameras are commonly used to record scene intensity in a computer memory or frame buffer. Scene data from the image sensor can then be analyzed with a wide variety of feature-detection techniques. Many algorithms found in textbooks on image processing make the implicit simplifying assumption of an ideal input image with clearly defined edges and uniform illumination. The ideal image model is helpful to aid the student in understanding the principles of operation, but when these algorithms are blindly applied to real-world images the results can be unsatisfactory. This paper examines some common measurement errors found in camera sensors and their underlying causes, and possible methods of error compensation. The role of the camera in a typical image-processing system is discussed, with emphasis on the origination of signal distortions. The effects of such things as lighting, optics, and sensor characteristics are considered. 11 refs., 7 figs.